2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
35 #include <sys/types.h>
38 #include <urcu/futex.h>
39 #include <urcu/uatomic.h>
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/compat/getenv.h>
49 #include <common/defaults.h>
50 #include <common/daemonize.h>
51 #include <common/futex.h>
52 #include <common/sessiond-comm/sessiond-comm.h>
53 #include <common/sessiond-comm/inet.h>
54 #include <common/sessiond-comm/relayd.h>
55 #include <common/uri.h>
56 #include <common/utils.h>
57 #include <common/align.h>
58 #include <common/config/session-config.h>
59 #include <common/dynamic-buffer.h>
60 #include <common/buffer-view.h>
61 #include <urcu/rculist.h>
64 #include "ctf-trace.h"
67 #include "lttng-relayd.h"
69 #include "health-relayd.h"
70 #include "testpoint.h"
71 #include "viewer-stream.h"
74 #include "connection.h"
75 #include "tracefile-array.h"
76 #include "tcp_keep_alive.h"
77 #include "sessiond-trace-chunks.h"
79 static const char *help_msg
=
80 #ifdef LTTNG_EMBED_HELP
81 #include <lttng-relayd.8.h>
87 enum relay_connection_status
{
88 RELAY_CONNECTION_STATUS_OK
,
89 /* An error occurred while processing an event on the connection. */
90 RELAY_CONNECTION_STATUS_ERROR
,
91 /* Connection closed/shutdown cleanly. */
92 RELAY_CONNECTION_STATUS_CLOSED
,
95 /* command line options */
96 char *opt_output_path
;
97 static int opt_daemon
, opt_background
;
100 * We need to wait for listener and live listener threads, as well as
101 * health check thread, before being ready to signal readiness.
103 #define NR_LTTNG_RELAY_READY 3
104 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
106 /* Size of receive buffer. */
107 #define RECV_DATA_BUFFER_SIZE 65536
108 #define FILE_COPY_BUFFER_SIZE 65536
110 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
111 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
113 static struct lttng_uri
*control_uri
;
114 static struct lttng_uri
*data_uri
;
115 static struct lttng_uri
*live_uri
;
117 const char *progname
;
119 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
120 static int tracing_group_name_override
;
122 const char * const config_section_name
= "relayd";
125 * Quit pipe for all threads. This permits a single cancellation point
126 * for all threads when receiving an event on the pipe.
128 int thread_quit_pipe
[2] = { -1, -1 };
131 * This pipe is used to inform the worker thread that a command is queued and
132 * ready to be processed.
134 static int relay_conn_pipe
[2] = { -1, -1 };
136 /* Shared between threads */
137 static int dispatch_thread_exit
;
139 static pthread_t listener_thread
;
140 static pthread_t dispatcher_thread
;
141 static pthread_t worker_thread
;
142 static pthread_t health_thread
;
145 * last_relay_stream_id_lock protects last_relay_stream_id increment
146 * atomicity on 32-bit architectures.
148 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
149 static uint64_t last_relay_stream_id
;
152 * Relay command queue.
154 * The relay_thread_listener and relay_thread_dispatcher communicate with this
157 static struct relay_conn_queue relay_conn_queue
;
159 /* Global relay stream hash table. */
160 struct lttng_ht
*relay_streams_ht
;
162 /* Global relay viewer stream hash table. */
163 struct lttng_ht
*viewer_streams_ht
;
165 /* Global relay sessions hash table. */
166 struct lttng_ht
*sessions_ht
;
168 /* Relayd health monitoring */
169 struct health_app
*health_relayd
;
171 struct sessiond_trace_chunk_registry
*sessiond_trace_chunk_registry
;
173 static struct option long_options
[] = {
174 { "control-port", 1, 0, 'C', },
175 { "data-port", 1, 0, 'D', },
176 { "live-port", 1, 0, 'L', },
177 { "daemonize", 0, 0, 'd', },
178 { "background", 0, 0, 'b', },
179 { "group", 1, 0, 'g', },
180 { "help", 0, 0, 'h', },
181 { "output", 1, 0, 'o', },
182 { "verbose", 0, 0, 'v', },
183 { "config", 1, 0, 'f' },
184 { "version", 0, 0, 'V' },
188 static const char *config_ignore_options
[] = { "help", "config", "version" };
191 * Take an option from the getopt output and set it in the right variable to be
194 * Return 0 on success else a negative value.
196 static int set_option(int opt
, const char *arg
, const char *optname
)
202 fprintf(stderr
, "option %s", optname
);
204 fprintf(stderr
, " with arg %s\n", arg
);
208 if (lttng_is_setuid_setgid()) {
209 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
210 "-C, --control-port");
212 ret
= uri_parse(arg
, &control_uri
);
214 ERR("Invalid control URI specified");
217 if (control_uri
->port
== 0) {
218 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
223 if (lttng_is_setuid_setgid()) {
224 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
227 ret
= uri_parse(arg
, &data_uri
);
229 ERR("Invalid data URI specified");
232 if (data_uri
->port
== 0) {
233 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
238 if (lttng_is_setuid_setgid()) {
239 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
242 ret
= uri_parse(arg
, &live_uri
);
244 ERR("Invalid live URI specified");
247 if (live_uri
->port
== 0) {
248 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
259 if (lttng_is_setuid_setgid()) {
260 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
263 tracing_group_name
= strdup(arg
);
264 if (tracing_group_name
== NULL
) {
269 tracing_group_name_override
= 1;
273 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
275 ERR("Cannot show --help for `lttng-relayd`");
280 fprintf(stdout
, "%s\n", VERSION
);
283 if (lttng_is_setuid_setgid()) {
284 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
287 ret
= asprintf(&opt_output_path
, "%s", arg
);
290 PERROR("asprintf opt_output_path");
296 /* Verbose level can increase using multiple -v */
298 lttng_opt_verbose
= config_parse_value(arg
);
300 /* Only 3 level of verbosity (-vvv). */
301 if (lttng_opt_verbose
< 3) {
302 lttng_opt_verbose
+= 1;
307 /* Unknown option or other error.
308 * Error is printed by getopt, just return */
321 * config_entry_handler_cb used to handle options read from a config file.
322 * See config_entry_handler_cb comment in common/config/session-config.h for the
323 * return value conventions.
325 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
329 if (!entry
|| !entry
->name
|| !entry
->value
) {
334 /* Check if the option is to be ignored */
335 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
336 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
341 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
342 /* Ignore if entry name is not fully matched. */
343 if (strcmp(entry
->name
, long_options
[i
].name
)) {
348 * If the option takes no argument on the command line,
349 * we have to check if the value is "true". We support
350 * non-zero numeric values, true, on and yes.
352 if (!long_options
[i
].has_arg
) {
353 ret
= config_parse_value(entry
->value
);
356 WARN("Invalid configuration value \"%s\" for option %s",
357 entry
->value
, entry
->name
);
359 /* False, skip boolean config option. */
364 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
368 WARN("Unrecognized option \"%s\" in daemon configuration file.",
375 static int set_options(int argc
, char **argv
)
377 int c
, ret
= 0, option_index
= 0, retval
= 0;
378 int orig_optopt
= optopt
, orig_optind
= optind
;
379 char *default_address
, *optstring
;
380 const char *config_path
= NULL
;
382 optstring
= utils_generate_optstring(long_options
,
383 sizeof(long_options
) / sizeof(struct option
));
389 /* Check for the --config option */
391 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
392 &option_index
)) != -1) {
396 } else if (c
!= 'f') {
400 if (lttng_is_setuid_setgid()) {
401 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
404 config_path
= utils_expand_path(optarg
);
406 ERR("Failed to resolve path: %s", optarg
);
411 ret
= config_get_section_entries(config_path
, config_section_name
,
412 config_entry_handler
, NULL
);
415 ERR("Invalid configuration option at line %i", ret
);
421 /* Reset getopt's global state */
422 optopt
= orig_optopt
;
423 optind
= orig_optind
;
425 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
430 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
437 /* assign default values */
438 if (control_uri
== NULL
) {
439 ret
= asprintf(&default_address
,
440 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
441 DEFAULT_NETWORK_CONTROL_PORT
);
443 PERROR("asprintf default data address");
448 ret
= uri_parse(default_address
, &control_uri
);
449 free(default_address
);
451 ERR("Invalid control URI specified");
456 if (data_uri
== NULL
) {
457 ret
= asprintf(&default_address
,
458 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
459 DEFAULT_NETWORK_DATA_PORT
);
461 PERROR("asprintf default data address");
466 ret
= uri_parse(default_address
, &data_uri
);
467 free(default_address
);
469 ERR("Invalid data URI specified");
474 if (live_uri
== NULL
) {
475 ret
= asprintf(&default_address
,
476 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
477 DEFAULT_NETWORK_VIEWER_PORT
);
479 PERROR("asprintf default viewer control address");
484 ret
= uri_parse(default_address
, &live_uri
);
485 free(default_address
);
487 ERR("Invalid viewer control URI specified");
498 static void print_global_objects(void)
500 rcu_register_thread();
502 print_viewer_streams();
503 print_relay_streams();
506 rcu_unregister_thread();
512 static void relayd_cleanup(void)
514 print_global_objects();
518 if (viewer_streams_ht
)
519 lttng_ht_destroy(viewer_streams_ht
);
520 if (relay_streams_ht
)
521 lttng_ht_destroy(relay_streams_ht
);
523 lttng_ht_destroy(sessions_ht
);
525 /* free the dynamically allocated opt_output_path */
526 free(opt_output_path
);
528 /* Close thread quit pipes */
529 utils_close_pipe(thread_quit_pipe
);
531 uri_free(control_uri
);
533 /* Live URI is freed in the live thread. */
535 if (tracing_group_name_override
) {
536 free((void *) tracing_group_name
);
541 * Write to writable pipe used to notify a thread.
543 static int notify_thread_pipe(int wpipe
)
547 ret
= lttng_write(wpipe
, "!", 1);
549 PERROR("write poll pipe");
557 static int notify_health_quit_pipe(int *pipe
)
561 ret
= lttng_write(pipe
[1], "4", 1);
563 PERROR("write relay health quit");
572 * Stop all relayd and relayd-live threads.
574 int lttng_relay_stop_threads(void)
578 /* Stopping all threads */
579 DBG("Terminating all threads");
580 if (notify_thread_pipe(thread_quit_pipe
[1])) {
581 ERR("write error on thread quit pipe");
585 if (notify_health_quit_pipe(health_quit_pipe
)) {
586 ERR("write error on health quit pipe");
589 /* Dispatch thread */
590 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
591 futex_nto1_wake(&relay_conn_queue
.futex
);
593 if (relayd_live_stop()) {
594 ERR("Error stopping live threads");
601 * Signal handler for the daemon
603 * Simply stop all worker threads, leaving main() return gracefully after
604 * joining all threads and calling cleanup().
606 static void sighandler(int sig
)
610 DBG("SIGINT caught");
611 if (lttng_relay_stop_threads()) {
612 ERR("Error stopping threads");
616 DBG("SIGTERM caught");
617 if (lttng_relay_stop_threads()) {
618 ERR("Error stopping threads");
622 CMM_STORE_SHARED(recv_child_signal
, 1);
630 * Setup signal handler for :
631 * SIGINT, SIGTERM, SIGPIPE
633 static int set_signal_handler(void)
639 if ((ret
= sigemptyset(&sigset
)) < 0) {
640 PERROR("sigemptyset");
647 sa
.sa_handler
= sighandler
;
648 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
653 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
658 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
663 sa
.sa_handler
= SIG_IGN
;
664 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
669 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
674 void lttng_relay_notify_ready(void)
676 /* Notify the parent of the fork() process that we are ready. */
677 if (opt_daemon
|| opt_background
) {
678 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
679 kill(child_ppid
, SIGUSR1
);
685 * Init thread quit pipe.
687 * Return -1 on error or 0 if all pipes are created.
689 static int init_thread_quit_pipe(void)
693 ret
= utils_create_pipe_cloexec(thread_quit_pipe
);
699 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
701 static int create_thread_poll_set(struct lttng_poll_event
*events
, int size
)
705 if (events
== NULL
|| size
== 0) {
710 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
716 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
728 * Check if the thread quit pipe was triggered.
730 * Return 1 if it was triggered else 0;
732 static int check_thread_quit_pipe(int fd
, uint32_t events
)
734 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
742 * Create and init socket from uri.
744 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
)
747 struct lttcomm_sock
*sock
= NULL
;
749 sock
= lttcomm_alloc_sock_from_uri(uri
);
751 ERR("Allocating socket");
755 ret
= lttcomm_create_sock(sock
);
759 DBG("Listening on sock %d", sock
->fd
);
761 ret
= sock
->ops
->bind(sock
);
763 PERROR("Failed to bind socket");
767 ret
= sock
->ops
->listen(sock
, -1);
777 lttcomm_destroy_sock(sock
);
783 * This thread manages the listening for new connections on the network
785 static void *relay_thread_listener(void *data
)
787 int i
, ret
, pollfd
, err
= -1;
788 uint32_t revents
, nb_fd
;
789 struct lttng_poll_event events
;
790 struct lttcomm_sock
*control_sock
, *data_sock
;
792 DBG("[thread] Relay listener started");
794 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
796 health_code_update();
798 control_sock
= relay_socket_create(control_uri
);
800 goto error_sock_control
;
803 data_sock
= relay_socket_create(data_uri
);
805 goto error_sock_relay
;
809 * Pass 3 as size here for the thread quit pipe, control and
812 ret
= create_thread_poll_set(&events
, 3);
814 goto error_create_poll
;
817 /* Add the control socket */
818 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
823 /* Add the data socket */
824 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
829 lttng_relay_notify_ready();
831 if (testpoint(relayd_thread_listener
)) {
832 goto error_testpoint
;
836 health_code_update();
838 DBG("Listener accepting connections");
842 ret
= lttng_poll_wait(&events
, -1);
846 * Restart interrupted system call.
848 if (errno
== EINTR
) {
856 DBG("Relay new connection received");
857 for (i
= 0; i
< nb_fd
; i
++) {
858 health_code_update();
860 /* Fetch once the poll data */
861 revents
= LTTNG_POLL_GETEV(&events
, i
);
862 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
864 /* Thread quit pipe has been closed. Killing thread. */
865 ret
= check_thread_quit_pipe(pollfd
, revents
);
871 if (revents
& LPOLLIN
) {
873 * A new connection is requested, therefore a
874 * sessiond/consumerd connection is allocated in
875 * this thread, enqueued to a global queue and
876 * dequeued (and freed) in the worker thread.
879 struct relay_connection
*new_conn
;
880 struct lttcomm_sock
*newsock
;
881 enum connection_type type
;
883 if (pollfd
== data_sock
->fd
) {
885 newsock
= data_sock
->ops
->accept(data_sock
);
886 DBG("Relay data connection accepted, socket %d",
889 assert(pollfd
== control_sock
->fd
);
890 type
= RELAY_CONTROL
;
891 newsock
= control_sock
->ops
->accept(control_sock
);
892 DBG("Relay control connection accepted, socket %d",
896 PERROR("accepting sock");
900 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
903 PERROR("setsockopt inet");
904 lttcomm_destroy_sock(newsock
);
908 ret
= socket_apply_keep_alive_config(newsock
->fd
);
910 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
912 lttcomm_destroy_sock(newsock
);
916 new_conn
= connection_create(newsock
, type
);
918 lttcomm_destroy_sock(newsock
);
922 /* Enqueue request for the dispatcher thread. */
923 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
927 * Wake the dispatch queue futex.
928 * Implicit memory barrier with the
929 * exchange in cds_wfcq_enqueue.
931 futex_nto1_wake(&relay_conn_queue
.futex
);
932 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
933 ERR("socket poll error");
936 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
946 lttng_poll_clean(&events
);
948 if (data_sock
->fd
>= 0) {
949 ret
= data_sock
->ops
->close(data_sock
);
954 lttcomm_destroy_sock(data_sock
);
956 if (control_sock
->fd
>= 0) {
957 ret
= control_sock
->ops
->close(control_sock
);
962 lttcomm_destroy_sock(control_sock
);
966 ERR("Health error occurred in %s", __func__
);
968 health_unregister(health_relayd
);
969 DBG("Relay listener thread cleanup complete");
970 lttng_relay_stop_threads();
975 * This thread manages the dispatching of the requests to worker threads
977 static void *relay_thread_dispatcher(void *data
)
981 struct cds_wfcq_node
*node
;
982 struct relay_connection
*new_conn
= NULL
;
984 DBG("[thread] Relay dispatcher started");
986 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
988 if (testpoint(relayd_thread_dispatcher
)) {
989 goto error_testpoint
;
992 health_code_update();
995 health_code_update();
997 /* Atomically prepare the queue futex */
998 futex_nto1_prepare(&relay_conn_queue
.futex
);
1000 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1005 health_code_update();
1007 /* Dequeue commands */
1008 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1009 &relay_conn_queue
.tail
);
1011 DBG("Woken up but nothing in the relay command queue");
1012 /* Continue thread execution */
1015 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1017 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1020 * Inform worker thread of the new request. This
1021 * call is blocking so we can be assured that
1022 * the data will be read at some point in time
1023 * or wait to the end of the world :)
1025 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1027 PERROR("write connection pipe");
1028 connection_put(new_conn
);
1031 } while (node
!= NULL
);
1033 /* Futex wait on queue. Blocking call on futex() */
1034 health_poll_entry();
1035 futex_nto1_wait(&relay_conn_queue
.futex
);
1039 /* Normal exit, no error */
1046 ERR("Health error occurred in %s", __func__
);
1048 health_unregister(health_relayd
);
1049 DBG("Dispatch thread dying");
1050 lttng_relay_stop_threads();
1055 * Set index data from the control port to a given index object.
1057 static int set_index_control_data(struct relay_index
*index
,
1058 struct lttcomm_relayd_index
*data
,
1059 struct relay_connection
*conn
)
1061 struct ctf_packet_index index_data
;
1064 * The index on disk is encoded in big endian.
1066 index_data
.packet_size
= htobe64(data
->packet_size
);
1067 index_data
.content_size
= htobe64(data
->content_size
);
1068 index_data
.timestamp_begin
= htobe64(data
->timestamp_begin
);
1069 index_data
.timestamp_end
= htobe64(data
->timestamp_end
);
1070 index_data
.events_discarded
= htobe64(data
->events_discarded
);
1071 index_data
.stream_id
= htobe64(data
->stream_id
);
1073 if (conn
->minor
>= 8) {
1074 index
->index_data
.stream_instance_id
= htobe64(data
->stream_instance_id
);
1075 index
->index_data
.packet_seq_num
= htobe64(data
->packet_seq_num
);
1078 return relay_index_set_data(index
, &index_data
);
1081 static bool session_streams_have_index(const struct relay_session
*session
)
1083 return session
->minor
>= 4 && !session
->snapshot
;
1087 * Handle the RELAYD_CREATE_SESSION command.
1089 * On success, send back the session id or else return a negative value.
1091 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1092 struct relay_connection
*conn
,
1093 const struct lttng_buffer_view
*payload
)
1097 struct relay_session
*session
= NULL
;
1098 struct lttcomm_relayd_status_session reply
;
1099 char session_name
[LTTNG_NAME_MAX
];
1100 char hostname
[LTTNG_HOST_NAME_MAX
];
1101 uint32_t live_timer
= 0;
1102 bool snapshot
= false;
1103 /* Left nil for peers < 2.11. */
1104 lttng_uuid sessiond_uuid
= {};
1106 memset(session_name
, 0, LTTNG_NAME_MAX
);
1107 memset(hostname
, 0, LTTNG_HOST_NAME_MAX
);
1109 memset(&reply
, 0, sizeof(reply
));
1111 if (conn
->minor
< 4) {
1112 /* From 2.1 to 2.3 */
1114 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1115 /* From 2.4 to 2.10 */
1116 ret
= cmd_create_session_2_4(payload
, session_name
,
1117 hostname
, &live_timer
, &snapshot
);
1119 /* From 2.11 to ... */
1120 ret
= cmd_create_session_2_11(payload
, session_name
,
1121 hostname
, &live_timer
, &snapshot
,
1123 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1124 /* The nil UUID is reserved for pre-2.11 clients. */
1125 ERR("Illegal nil UUID announced by peer in create session command");
1135 session
= session_create(session_name
, hostname
, live_timer
,
1136 snapshot
, sessiond_uuid
, conn
->major
, conn
->minor
);
1141 assert(!conn
->session
);
1142 conn
->session
= session
;
1143 DBG("Created session %" PRIu64
, session
->id
);
1145 reply
.session_id
= htobe64(session
->id
);
1149 reply
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1151 reply
.ret_code
= htobe32(LTTNG_OK
);
1154 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1155 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1156 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1160 if (ret
< 0 && session
) {
1161 session_put(session
);
1167 * When we have received all the streams and the metadata for a channel,
1168 * we make them visible to the viewer threads.
1170 static void publish_connection_local_streams(struct relay_connection
*conn
)
1172 struct relay_stream
*stream
;
1173 struct relay_session
*session
= conn
->session
;
1176 * We publish all streams belonging to a session atomically wrt
1179 pthread_mutex_lock(&session
->lock
);
1181 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1183 stream_publish(stream
);
1188 * Inform the viewer that there are new streams in the session.
1190 if (session
->viewer_attached
) {
1191 uatomic_set(&session
->new_streams
, 1);
1193 pthread_mutex_unlock(&session
->lock
);
1197 * relay_add_stream: allocate a new stream for a session
1199 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1200 struct relay_connection
*conn
,
1201 const struct lttng_buffer_view
*payload
)
1205 struct relay_session
*session
= conn
->session
;
1206 struct relay_stream
*stream
= NULL
;
1207 struct lttcomm_relayd_status_stream reply
;
1208 struct ctf_trace
*trace
= NULL
;
1209 uint64_t stream_handle
= -1ULL;
1210 char *path_name
= NULL
, *channel_name
= NULL
;
1211 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1212 struct relay_stream_chunk_id stream_chunk_id
= { 0 };
1214 if (!session
|| !conn
->version_check_done
) {
1215 ERR("Trying to add a stream before version check");
1217 goto end_no_session
;
1220 if (session
->minor
== 1) {
1222 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1224 } else if (session
->minor
> 1 && session
->minor
< 11) {
1225 /* From 2.2 to 2.10 */
1226 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1227 &channel_name
, &tracefile_size
, &tracefile_count
);
1229 /* From 2.11 to ... */
1230 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1231 &channel_name
, &tracefile_size
, &tracefile_count
,
1232 &stream_chunk_id
.value
);
1233 stream_chunk_id
.is_set
= true;
1240 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1244 /* This stream here has one reference on the trace. */
1246 pthread_mutex_lock(&last_relay_stream_id_lock
);
1247 stream_handle
= ++last_relay_stream_id
;
1248 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1250 /* We pass ownership of path_name and channel_name. */
1251 stream
= stream_create(trace
, stream_handle
, path_name
,
1252 channel_name
, tracefile_size
, tracefile_count
,
1255 channel_name
= NULL
;
1258 * Streams are the owners of their trace. Reference to trace is
1259 * kept within stream_create().
1261 ctf_trace_put(trace
);
1264 memset(&reply
, 0, sizeof(reply
));
1265 reply
.handle
= htobe64(stream_handle
);
1267 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1269 reply
.ret_code
= htobe32(LTTNG_OK
);
1272 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1273 sizeof(struct lttcomm_relayd_status_stream
), 0);
1274 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1275 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1287 * relay_close_stream: close a specific stream
1289 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1290 struct relay_connection
*conn
,
1291 const struct lttng_buffer_view
*payload
)
1295 struct relay_session
*session
= conn
->session
;
1296 struct lttcomm_relayd_close_stream stream_info
;
1297 struct lttcomm_relayd_generic_reply reply
;
1298 struct relay_stream
*stream
;
1300 DBG("Close stream received");
1302 if (!session
|| !conn
->version_check_done
) {
1303 ERR("Trying to close a stream before version check");
1305 goto end_no_session
;
1308 if (payload
->size
< sizeof(stream_info
)) {
1309 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1310 sizeof(stream_info
), payload
->size
);
1312 goto end_no_session
;
1314 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1315 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1316 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1318 stream
= stream_get_by_id(stream_info
.stream_id
);
1325 * Set last_net_seq_num before the close flag. Required by data
1328 pthread_mutex_lock(&stream
->lock
);
1329 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1330 pthread_mutex_unlock(&stream
->lock
);
1333 * This is one of the conditions which may trigger a stream close
1334 * with the others being:
1335 * 1) A close command is received for a stream
1336 * 2) The control connection owning the stream is closed
1337 * 3) We have received all of the stream's data _after_ a close
1340 try_stream_close(stream
);
1341 if (stream
->is_metadata
) {
1342 struct relay_viewer_stream
*vstream
;
1344 vstream
= viewer_stream_get_by_id(stream
->stream_handle
);
1346 if (vstream
->metadata_sent
== stream
->metadata_received
) {
1348 * Since all the metadata has been sent to the
1349 * viewer and that we have a request to close
1350 * its stream, we can safely teardown the
1351 * corresponding metadata viewer stream.
1353 viewer_stream_put(vstream
);
1355 /* Put local reference. */
1356 viewer_stream_put(vstream
);
1363 memset(&reply
, 0, sizeof(reply
));
1365 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1367 reply
.ret_code
= htobe32(LTTNG_OK
);
1369 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1370 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1371 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1372 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1382 * relay_reset_metadata: reset a metadata stream
1385 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1386 struct relay_connection
*conn
,
1387 const struct lttng_buffer_view
*payload
)
1391 struct relay_session
*session
= conn
->session
;
1392 struct lttcomm_relayd_reset_metadata stream_info
;
1393 struct lttcomm_relayd_generic_reply reply
;
1394 struct relay_stream
*stream
;
1396 DBG("Reset metadata received");
1398 if (!session
|| !conn
->version_check_done
) {
1399 ERR("Trying to reset a metadata stream before version check");
1401 goto end_no_session
;
1404 if (payload
->size
< sizeof(stream_info
)) {
1405 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1406 sizeof(stream_info
), payload
->size
);
1408 goto end_no_session
;
1410 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1411 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1412 stream_info
.version
= be64toh(stream_info
.version
);
1414 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1416 /* Unsupported for live sessions for now. */
1417 if (session
->live_timer
!= 0) {
1422 stream
= stream_get_by_id(stream_info
.stream_id
);
1427 pthread_mutex_lock(&stream
->lock
);
1428 if (!stream
->is_metadata
) {
1433 ret
= utils_rotate_stream_file(stream
->path_name
, stream
->channel_name
,
1434 0, 0, -1, -1, stream
->stream_fd
->fd
, NULL
,
1435 &stream
->stream_fd
->fd
);
1437 ERR("Failed to rotate metadata file %s of channel %s",
1438 stream
->path_name
, stream
->channel_name
);
1443 pthread_mutex_unlock(&stream
->lock
);
1447 memset(&reply
, 0, sizeof(reply
));
1449 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1451 reply
.ret_code
= htobe32(LTTNG_OK
);
1453 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1454 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1455 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1456 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1466 * relay_unknown_command: send -1 if received unknown command
1468 static void relay_unknown_command(struct relay_connection
*conn
)
1470 struct lttcomm_relayd_generic_reply reply
;
1473 memset(&reply
, 0, sizeof(reply
));
1474 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1475 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1476 if (send_ret
< sizeof(reply
)) {
1477 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1482 * relay_start: send an acknowledgment to the client to tell if we are
1483 * ready to receive data. We are ready if a session is established.
1485 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1486 struct relay_connection
*conn
,
1487 const struct lttng_buffer_view
*payload
)
1491 struct lttcomm_relayd_generic_reply reply
;
1492 struct relay_session
*session
= conn
->session
;
1495 DBG("Trying to start the streaming without a session established");
1496 ret
= htobe32(LTTNG_ERR_UNK
);
1499 memset(&reply
, 0, sizeof(reply
));
1500 reply
.ret_code
= htobe32(LTTNG_OK
);
1501 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1503 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1504 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1513 * Append padding to the file pointed by the file descriptor fd.
1515 static int write_padding_to_file(int fd
, uint32_t size
)
1524 zeros
= zmalloc(size
);
1525 if (zeros
== NULL
) {
1526 PERROR("zmalloc zeros for padding");
1531 ret
= lttng_write(fd
, zeros
, size
);
1533 PERROR("write padding to file");
1543 * Close the current index file if it is open, and create a new one.
1545 * Return 0 on success, -1 on error.
1548 int create_rotate_index_file(struct relay_stream
*stream
,
1549 const char *stream_path
)
1552 uint32_t major
, minor
;
1554 /* Put ref on previous index_file. */
1555 if (stream
->index_file
) {
1556 lttng_index_file_put(stream
->index_file
);
1557 stream
->index_file
= NULL
;
1559 major
= stream
->trace
->session
->major
;
1560 minor
= stream
->trace
->session
->minor
;
1561 stream
->index_file
= lttng_index_file_create(stream_path
,
1562 stream
->channel_name
,
1563 -1, -1, stream
->tracefile_size
,
1564 tracefile_array_get_file_index_head(stream
->tfa
),
1565 lttng_to_index_major(major
, minor
),
1566 lttng_to_index_minor(major
, minor
));
1567 if (!stream
->index_file
) {
1579 int do_rotate_stream_data(struct relay_stream
*stream
)
1583 DBG("Rotating stream %" PRIu64
" data file",
1584 stream
->stream_handle
);
1585 /* Perform the stream rotation. */
1586 ret
= utils_rotate_stream_file(stream
->path_name
,
1587 stream
->channel_name
, stream
->tracefile_size
,
1588 stream
->tracefile_count
, -1,
1589 -1, stream
->stream_fd
->fd
,
1590 NULL
, &stream
->stream_fd
->fd
);
1592 ERR("Rotating stream output file");
1595 stream
->tracefile_size_current
= 0;
1596 stream
->pos_after_last_complete_data_index
= 0;
1597 stream
->data_rotated
= true;
1599 if (stream
->data_rotated
&& stream
->index_rotated
) {
1600 /* Rotation completed; reset its state. */
1601 DBG("Rotation completed for stream %" PRIu64
,
1602 stream
->stream_handle
);
1603 stream
->rotate_at_seq_num
= -1ULL;
1604 stream
->data_rotated
= false;
1605 stream
->index_rotated
= false;
1612 * If too much data has been written in a tracefile before we received the
1613 * rotation command, we have to move the excess data to the new tracefile and
1614 * perform the rotation. This can happen because the control and data
1615 * connections are separate, the indexes as well as the commands arrive from
1616 * the control connection and we have no control over the order so we could be
1617 * in a situation where too much data has been received on the data connection
1618 * before the rotation command on the control connection arrives.
1621 int rotate_truncate_stream(struct relay_stream
*stream
)
1625 uint64_t diff
, pos
= 0;
1626 char buf
[FILE_COPY_BUFFER_SIZE
];
1628 assert(!stream
->is_metadata
);
1630 assert(stream
->tracefile_size_current
>
1631 stream
->pos_after_last_complete_data_index
);
1632 diff
= stream
->tracefile_size_current
-
1633 stream
->pos_after_last_complete_data_index
;
1635 /* Create the new tracefile. */
1636 new_fd
= utils_create_stream_file(stream
->path_name
,
1637 stream
->channel_name
,
1638 stream
->tracefile_size
, stream
->tracefile_count
,
1639 /* uid */ -1, /* gid */ -1, /* suffix */ NULL
);
1641 ERR("Failed to create new stream file at path %s for channel %s",
1642 stream
->path_name
, stream
->channel_name
);
1648 * Rewind the current tracefile to the position at which the rotation
1649 * should have occurred.
1651 lseek_ret
= lseek(stream
->stream_fd
->fd
,
1652 stream
->pos_after_last_complete_data_index
, SEEK_SET
);
1653 if (lseek_ret
< 0) {
1654 PERROR("seek truncate stream");
1659 /* Move data from the old file to the new file. */
1660 while (pos
< diff
) {
1661 uint64_t count
, bytes_left
;
1664 bytes_left
= diff
- pos
;
1665 count
= bytes_left
> sizeof(buf
) ? sizeof(buf
) : bytes_left
;
1666 assert(count
<= SIZE_MAX
);
1668 io_ret
= lttng_read(stream
->stream_fd
->fd
, buf
, count
);
1669 if (io_ret
< (ssize_t
) count
) {
1670 char error_string
[256];
1672 snprintf(error_string
, sizeof(error_string
),
1673 "Failed to read %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1674 count
, stream
->stream_fd
->fd
, io_ret
);
1676 PERROR("%s", error_string
);
1678 ERR("%s", error_string
);
1684 io_ret
= lttng_write(new_fd
, buf
, count
);
1685 if (io_ret
< (ssize_t
) count
) {
1686 char error_string
[256];
1688 snprintf(error_string
, sizeof(error_string
),
1689 "Failed to write %" PRIu64
" bytes from fd %i in rotate_truncate_stream(), returned %zi",
1690 count
, new_fd
, io_ret
);
1692 PERROR("%s", error_string
);
1694 ERR("%s", error_string
);
1703 /* Truncate the file to get rid of the excess data. */
1704 ret
= ftruncate(stream
->stream_fd
->fd
,
1705 stream
->pos_after_last_complete_data_index
);
1707 PERROR("ftruncate");
1711 ret
= close(stream
->stream_fd
->fd
);
1713 PERROR("Closing tracefile");
1718 * Update the offset and FD of all the eventual indexes created by the
1719 * data connection before the rotation command arrived.
1721 ret
= relay_index_switch_all_files(stream
);
1723 ERR("Failed to rotate index file");
1727 stream
->stream_fd
->fd
= new_fd
;
1728 stream
->tracefile_size_current
= diff
;
1729 stream
->pos_after_last_complete_data_index
= 0;
1730 stream
->rotate_at_seq_num
= -1ULL;
1739 * Check if a stream's index file should be rotated (for session rotation).
1740 * Must be called with the stream lock held.
1742 * Return 0 on success, a negative value on error.
1745 int try_rotate_stream_index(struct relay_stream
*stream
)
1749 if (stream
->rotate_at_seq_num
== -1ULL) {
1750 /* No rotation expected. */
1754 if (stream
->index_rotated
) {
1755 /* Rotation of the index has already occurred. */
1759 if (stream
->prev_index_seq
== -1ULL ||
1760 stream
->prev_index_seq
< stream
->rotate_at_seq_num
) {
1761 DBG("Stream %" PRIu64
" index not yet ready for rotation (rotate_at_seq_num = %" PRIu64
", prev_index_seq = %" PRIu64
")",
1762 stream
->stream_handle
,
1763 stream
->rotate_at_seq_num
,
1764 stream
->prev_index_seq
);
1766 } else if (stream
->prev_index_seq
!= stream
->rotate_at_seq_num
) {
1768 * Unexpected, protocol error/bug.
1769 * It could mean that we received a rotation position
1770 * that is in the past.
1772 ERR("Stream %" PRIu64
" index is in an inconsistent state (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
", prev_index_seq = %" PRIu64
")",
1773 stream
->stream_handle
,
1774 stream
->rotate_at_seq_num
,
1775 stream
->prev_data_seq
,
1776 stream
->prev_index_seq
);
1780 DBG("Rotating stream %" PRIu64
" index file",
1781 stream
->stream_handle
);
1782 ret
= create_rotate_index_file(stream
, stream
->path_name
);
1783 stream
->index_rotated
= true;
1785 if (stream
->data_rotated
&& stream
->index_rotated
) {
1786 /* Rotation completed; reset its state. */
1787 DBG("Rotation completed for stream %" PRIu64
,
1788 stream
->stream_handle
);
1789 stream
->rotate_at_seq_num
= -1ULL;
1790 stream
->data_rotated
= false;
1791 stream
->index_rotated
= false;
1800 * Check if a stream's data file (as opposed to index) should be rotated
1801 * (for session rotation).
1802 * Must be called with the stream lock held.
1804 * Return 0 on success, a negative value on error.
1807 int try_rotate_stream_data(struct relay_stream
*stream
)
1811 if (stream
->rotate_at_seq_num
== -1ULL) {
1812 /* No rotation expected. */
1816 if (stream
->data_rotated
) {
1817 /* Rotation of the data file has already occurred. */
1821 if (stream
->prev_data_seq
== -1ULL ||
1822 stream
->prev_data_seq
< stream
->rotate_at_seq_num
) {
1823 DBG("Stream %" PRIu64
" not yet ready for rotation (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
")",
1824 stream
->stream_handle
,
1825 stream
->rotate_at_seq_num
,
1826 stream
->prev_data_seq
);
1828 } else if (stream
->prev_data_seq
> stream
->rotate_at_seq_num
) {
1830 * prev_data_seq is checked here since indexes and rotation
1831 * commands are serialized with respect to each other.
1833 DBG("Rotation after too much data has been written in tracefile "
1834 "for stream %" PRIu64
", need to truncate before "
1835 "rotating", stream
->stream_handle
);
1836 ret
= rotate_truncate_stream(stream
);
1838 ERR("Failed to truncate stream");
1841 } else if (stream
->prev_data_seq
!= stream
->rotate_at_seq_num
) {
1843 * Unexpected, protocol error/bug.
1844 * It could mean that we received a rotation position
1845 * that is in the past.
1847 ERR("Stream %" PRIu64
" data is in an inconsistent state (rotate_at_seq_num = %" PRIu64
", prev_data_seq = %" PRIu64
")",
1848 stream
->stream_handle
,
1849 stream
->rotate_at_seq_num
,
1850 stream
->prev_data_seq
);
1854 ret
= do_rotate_stream_data(stream
);
1862 * relay_recv_metadata: receive the metadata for the session.
1864 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1865 struct relay_connection
*conn
,
1866 const struct lttng_buffer_view
*payload
)
1870 struct relay_session
*session
= conn
->session
;
1871 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1872 struct relay_stream
*metadata_stream
;
1873 uint64_t metadata_payload_size
;
1876 ERR("Metadata sent before version check");
1881 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1882 ERR("Incorrect data size");
1886 metadata_payload_size
= recv_hdr
->data_size
-
1887 sizeof(struct lttcomm_relayd_metadata_payload
);
1889 memcpy(&metadata_payload_header
, payload
->data
,
1890 sizeof(metadata_payload_header
));
1891 metadata_payload_header
.stream_id
= be64toh(
1892 metadata_payload_header
.stream_id
);
1893 metadata_payload_header
.padding_size
= be32toh(
1894 metadata_payload_header
.padding_size
);
1896 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1897 if (!metadata_stream
) {
1902 pthread_mutex_lock(&metadata_stream
->lock
);
1904 size_ret
= lttng_write(metadata_stream
->stream_fd
->fd
,
1905 payload
->data
+ sizeof(metadata_payload_header
),
1906 metadata_payload_size
);
1907 if (size_ret
< metadata_payload_size
) {
1908 ERR("Relay error writing metadata on file");
1913 size_ret
= write_padding_to_file(metadata_stream
->stream_fd
->fd
,
1914 metadata_payload_header
.padding_size
);
1915 if (size_ret
< (int64_t) metadata_payload_header
.padding_size
) {
1920 metadata_stream
->metadata_received
+=
1921 metadata_payload_size
+ metadata_payload_header
.padding_size
;
1922 DBG2("Relay metadata written. Updated metadata_received %" PRIu64
,
1923 metadata_stream
->metadata_received
);
1925 ret
= try_rotate_stream_data(metadata_stream
);
1931 pthread_mutex_unlock(&metadata_stream
->lock
);
1932 stream_put(metadata_stream
);
1938 * relay_send_version: send relayd version number
1940 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1941 struct relay_connection
*conn
,
1942 const struct lttng_buffer_view
*payload
)
1946 struct lttcomm_relayd_version reply
, msg
;
1947 bool compatible
= true;
1949 conn
->version_check_done
= true;
1951 /* Get version from the other side. */
1952 if (payload
->size
< sizeof(msg
)) {
1953 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1954 sizeof(msg
), payload
->size
);
1959 memcpy(&msg
, payload
->data
, sizeof(msg
));
1960 msg
.major
= be32toh(msg
.major
);
1961 msg
.minor
= be32toh(msg
.minor
);
1963 memset(&reply
, 0, sizeof(reply
));
1964 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1965 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1967 /* Major versions must be the same */
1968 if (reply
.major
!= msg
.major
) {
1969 DBG("Incompatible major versions (%u vs %u), deleting session",
1970 reply
.major
, msg
.major
);
1974 conn
->major
= reply
.major
;
1975 /* We adapt to the lowest compatible version */
1976 if (reply
.minor
<= msg
.minor
) {
1977 conn
->minor
= reply
.minor
;
1979 conn
->minor
= msg
.minor
;
1982 reply
.major
= htobe32(reply
.major
);
1983 reply
.minor
= htobe32(reply
.minor
);
1984 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1986 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1987 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2000 DBG("Version check done using protocol %u.%u", conn
->major
,
2008 * Check for data pending for a given stream id from the session daemon.
2010 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2011 struct relay_connection
*conn
,
2012 const struct lttng_buffer_view
*payload
)
2014 struct relay_session
*session
= conn
->session
;
2015 struct lttcomm_relayd_data_pending msg
;
2016 struct lttcomm_relayd_generic_reply reply
;
2017 struct relay_stream
*stream
;
2020 uint64_t stream_seq
;
2022 DBG("Data pending command received");
2024 if (!session
|| !conn
->version_check_done
) {
2025 ERR("Trying to check for data before version check");
2027 goto end_no_session
;
2030 if (payload
->size
< sizeof(msg
)) {
2031 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2032 sizeof(msg
), payload
->size
);
2034 goto end_no_session
;
2036 memcpy(&msg
, payload
->data
, sizeof(msg
));
2037 msg
.stream_id
= be64toh(msg
.stream_id
);
2038 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2040 stream
= stream_get_by_id(msg
.stream_id
);
2041 if (stream
== NULL
) {
2046 pthread_mutex_lock(&stream
->lock
);
2048 if (session_streams_have_index(session
)) {
2050 * Ensure that both the index and stream data have been
2051 * flushed up to the requested point.
2053 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2055 stream_seq
= stream
->prev_data_seq
;
2057 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
2058 ", prev_index_seq %" PRIu64
2059 ", and last_seq %" PRIu64
, msg
.stream_id
,
2060 stream
->prev_data_seq
, stream
->prev_index_seq
,
2061 msg
.last_net_seq_num
);
2063 /* Avoid wrapping issue */
2064 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
2065 /* Data has in fact been written and is NOT pending */
2068 /* Data still being streamed thus pending */
2072 stream
->data_pending_check_done
= true;
2073 pthread_mutex_unlock(&stream
->lock
);
2078 memset(&reply
, 0, sizeof(reply
));
2079 reply
.ret_code
= htobe32(ret
);
2080 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2081 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2082 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2092 * Wait for the control socket to reach a quiescent state.
2094 * Note that for now, when receiving this command from the session
2095 * daemon, this means that every subsequent commands or data received on
2096 * the control socket has been handled. So, this is why we simply return
2099 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
2100 struct relay_connection
*conn
,
2101 const struct lttng_buffer_view
*payload
)
2105 struct relay_stream
*stream
;
2106 struct lttcomm_relayd_quiescent_control msg
;
2107 struct lttcomm_relayd_generic_reply reply
;
2109 DBG("Checking quiescent state on control socket");
2111 if (!conn
->session
|| !conn
->version_check_done
) {
2112 ERR("Trying to check for data before version check");
2114 goto end_no_session
;
2117 if (payload
->size
< sizeof(msg
)) {
2118 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2119 sizeof(msg
), payload
->size
);
2121 goto end_no_session
;
2123 memcpy(&msg
, payload
->data
, sizeof(msg
));
2124 msg
.stream_id
= be64toh(msg
.stream_id
);
2126 stream
= stream_get_by_id(msg
.stream_id
);
2130 pthread_mutex_lock(&stream
->lock
);
2131 stream
->data_pending_check_done
= true;
2132 pthread_mutex_unlock(&stream
->lock
);
2134 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2137 memset(&reply
, 0, sizeof(reply
));
2138 reply
.ret_code
= htobe32(LTTNG_OK
);
2139 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2140 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2141 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2153 * Initialize a data pending command. This means that a consumer is about
2154 * to ask for data pending for each stream it holds. Simply iterate over
2155 * all streams of a session and set the data_pending_check_done flag.
2157 * This command returns to the client a LTTNG_OK code.
2159 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2160 struct relay_connection
*conn
,
2161 const struct lttng_buffer_view
*payload
)
2165 struct lttng_ht_iter iter
;
2166 struct lttcomm_relayd_begin_data_pending msg
;
2167 struct lttcomm_relayd_generic_reply reply
;
2168 struct relay_stream
*stream
;
2173 DBG("Init streams for data pending");
2175 if (!conn
->session
|| !conn
->version_check_done
) {
2176 ERR("Trying to check for data before version check");
2178 goto end_no_session
;
2181 if (payload
->size
< sizeof(msg
)) {
2182 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2183 sizeof(msg
), payload
->size
);
2185 goto end_no_session
;
2187 memcpy(&msg
, payload
->data
, sizeof(msg
));
2188 msg
.session_id
= be64toh(msg
.session_id
);
2191 * Iterate over all streams to set the begin data pending flag.
2192 * For now, the streams are indexed by stream handle so we have
2193 * to iterate over all streams to find the one associated with
2194 * the right session_id.
2197 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2199 if (!stream_get(stream
)) {
2202 if (stream
->trace
->session
->id
== msg
.session_id
) {
2203 pthread_mutex_lock(&stream
->lock
);
2204 stream
->data_pending_check_done
= false;
2205 pthread_mutex_unlock(&stream
->lock
);
2206 DBG("Set begin data pending flag to stream %" PRIu64
,
2207 stream
->stream_handle
);
2213 memset(&reply
, 0, sizeof(reply
));
2214 /* All good, send back reply. */
2215 reply
.ret_code
= htobe32(LTTNG_OK
);
2217 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2218 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2219 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2231 * End data pending command. This will check, for a given session id, if
2232 * each stream associated with it has its data_pending_check_done flag
2233 * set. If not, this means that the client lost track of the stream but
2234 * the data is still being streamed on our side. In this case, we inform
2235 * the client that data is in flight.
2237 * Return to the client if there is data in flight or not with a ret_code.
2239 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2240 struct relay_connection
*conn
,
2241 const struct lttng_buffer_view
*payload
)
2245 struct lttng_ht_iter iter
;
2246 struct lttcomm_relayd_end_data_pending msg
;
2247 struct lttcomm_relayd_generic_reply reply
;
2248 struct relay_stream
*stream
;
2249 uint32_t is_data_inflight
= 0;
2251 DBG("End data pending command");
2253 if (!conn
->session
|| !conn
->version_check_done
) {
2254 ERR("Trying to check for data before version check");
2256 goto end_no_session
;
2259 if (payload
->size
< sizeof(msg
)) {
2260 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2261 sizeof(msg
), payload
->size
);
2263 goto end_no_session
;
2265 memcpy(&msg
, payload
->data
, sizeof(msg
));
2266 msg
.session_id
= be64toh(msg
.session_id
);
2269 * Iterate over all streams to see if the begin data pending
2273 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2275 if (!stream_get(stream
)) {
2278 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2282 pthread_mutex_lock(&stream
->lock
);
2283 if (!stream
->data_pending_check_done
) {
2284 uint64_t stream_seq
;
2286 if (session_streams_have_index(conn
->session
)) {
2288 * Ensure that both the index and stream data have been
2289 * flushed up to the requested point.
2291 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2293 stream_seq
= stream
->prev_data_seq
;
2295 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
2296 is_data_inflight
= 1;
2297 DBG("Data is still in flight for stream %" PRIu64
,
2298 stream
->stream_handle
);
2299 pthread_mutex_unlock(&stream
->lock
);
2304 pthread_mutex_unlock(&stream
->lock
);
2309 memset(&reply
, 0, sizeof(reply
));
2310 /* All good, send back reply. */
2311 reply
.ret_code
= htobe32(is_data_inflight
);
2313 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2314 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2315 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2327 * Receive an index for a specific stream.
2329 * Return 0 on success else a negative value.
2331 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2332 struct relay_connection
*conn
,
2333 const struct lttng_buffer_view
*payload
)
2337 struct relay_session
*session
= conn
->session
;
2338 struct lttcomm_relayd_index index_info
;
2339 struct relay_index
*index
;
2340 struct lttcomm_relayd_generic_reply reply
;
2341 struct relay_stream
*stream
;
2346 DBG("Relay receiving index");
2348 if (!session
|| !conn
->version_check_done
) {
2349 ERR("Trying to close a stream before version check");
2351 goto end_no_session
;
2354 msg_len
= lttcomm_relayd_index_len(
2355 lttng_to_index_major(conn
->major
, conn
->minor
),
2356 lttng_to_index_minor(conn
->major
, conn
->minor
));
2357 if (payload
->size
< msg_len
) {
2358 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2359 msg_len
, payload
->size
);
2361 goto end_no_session
;
2363 memcpy(&index_info
, payload
->data
, msg_len
);
2364 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2365 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2366 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2367 index_info
.content_size
= be64toh(index_info
.content_size
);
2368 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2369 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2370 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2371 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2373 if (conn
->minor
>= 8) {
2374 index_info
.stream_instance_id
=
2375 be64toh(index_info
.stream_instance_id
);
2376 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2379 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2381 ERR("stream_get_by_id not found");
2385 pthread_mutex_lock(&stream
->lock
);
2387 /* Live beacon handling */
2388 if (index_info
.packet_size
== 0) {
2389 DBG("Received live beacon for stream %" PRIu64
,
2390 stream
->stream_handle
);
2393 * Only flag a stream inactive when it has already
2394 * received data and no indexes are in flight.
2396 if (stream
->index_received_seqcount
> 0
2397 && stream
->indexes_in_flight
== 0) {
2398 stream
->beacon_ts_end
= index_info
.timestamp_end
;
2401 goto end_stream_put
;
2403 stream
->beacon_ts_end
= -1ULL;
2406 if (stream
->ctf_stream_id
== -1ULL) {
2407 stream
->ctf_stream_id
= index_info
.stream_id
;
2409 index
= relay_index_get_by_id_or_create(stream
, index_info
.net_seq_num
);
2412 ERR("relay_index_get_by_id_or_create index NULL");
2413 goto end_stream_put
;
2415 if (set_index_control_data(index
, &index_info
, conn
)) {
2416 ERR("set_index_control_data error");
2417 relay_index_put(index
);
2419 goto end_stream_put
;
2421 ret
= relay_index_try_flush(index
);
2423 tracefile_array_commit_seq(stream
->tfa
);
2424 stream
->index_received_seqcount
++;
2425 stream
->pos_after_last_complete_data_index
+= index
->total_size
;
2426 stream
->prev_index_seq
= index_info
.net_seq_num
;
2428 ret
= try_rotate_stream_index(stream
);
2430 goto end_stream_put
;
2432 } else if (ret
> 0) {
2439 * relay_index_try_flush is responsible for the self-reference
2440 * put of the index object on error.
2442 ERR("relay_index_try_flush error %d", ret
);
2447 pthread_mutex_unlock(&stream
->lock
);
2452 memset(&reply
, 0, sizeof(reply
));
2454 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2456 reply
.ret_code
= htobe32(LTTNG_OK
);
2458 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2459 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2460 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2469 * Receive the streams_sent message.
2471 * Return 0 on success else a negative value.
2473 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2474 struct relay_connection
*conn
,
2475 const struct lttng_buffer_view
*payload
)
2479 struct lttcomm_relayd_generic_reply reply
;
2483 DBG("Relay receiving streams_sent");
2485 if (!conn
->session
|| !conn
->version_check_done
) {
2486 ERR("Trying to close a stream before version check");
2488 goto end_no_session
;
2492 * Publish every pending stream in the connection recv list which are
2493 * now ready to be used by the viewer.
2495 publish_connection_local_streams(conn
);
2497 memset(&reply
, 0, sizeof(reply
));
2498 reply
.ret_code
= htobe32(LTTNG_OK
);
2499 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2500 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2501 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2514 * relay_rotate_session_stream: rotate a stream to a new tracefile for the session
2515 * rotation feature (not the tracefile rotation feature).
2517 static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
2518 struct relay_connection
*conn
,
2519 const struct lttng_buffer_view
*payload
)
2523 struct relay_session
*session
= conn
->session
;
2524 struct lttcomm_relayd_rotate_stream stream_info
;
2525 struct lttcomm_relayd_generic_reply reply
;
2526 struct relay_stream
*stream
;
2529 struct lttng_buffer_view new_path_view
;
2531 DBG("Rotate stream received");
2533 if (!session
|| !conn
->version_check_done
) {
2534 ERR("Trying to rotate a stream before version check");
2539 if (session
->major
== 2 && session
->minor
< 11) {
2540 ERR("Unsupported feature before 2.11");
2545 header_len
= sizeof(struct lttcomm_relayd_rotate_stream
);
2547 if (payload
->size
< header_len
) {
2548 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2549 header_len
, payload
->size
);
2554 memcpy(&stream_info
, payload
->data
, header_len
);
2556 /* Convert to host */
2557 stream_info
.pathname_length
= be32toh(stream_info
.pathname_length
);
2558 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
2559 stream_info
.new_chunk_id
= be64toh(stream_info
.new_chunk_id
);
2560 stream_info
.rotate_at_seq_num
= be64toh(stream_info
.rotate_at_seq_num
);
2562 path_len
= stream_info
.pathname_length
;
2563 if (payload
->size
< header_len
+ path_len
) {
2564 ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes",
2565 header_len
+ path_len
, payload
->size
);
2570 /* Ensure it fits in local filename length. */
2571 if (path_len
>= LTTNG_PATH_MAX
) {
2572 ret
= -ENAMETOOLONG
;
2573 ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes",
2574 path_len
, LTTNG_PATH_MAX
);
2578 new_path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2579 stream_info
.pathname_length
);
2581 stream
= stream_get_by_id(stream_info
.stream_id
);
2587 pthread_mutex_lock(&stream
->lock
);
2590 * Update the trace path (just the folder, the stream name does not
2593 free(stream
->prev_path_name
);
2594 stream
->prev_path_name
= stream
->path_name
;
2595 stream
->path_name
= create_output_path(new_path_view
.data
);
2596 if (!stream
->path_name
) {
2597 ERR("Failed to create a new output path");
2599 goto end_stream_unlock
;
2601 ret
= utils_mkdir_recursive(stream
->path_name
, S_IRWXU
| S_IRWXG
,
2604 ERR("relay creating output directory");
2606 goto end_stream_unlock
;
2609 assert(stream
->current_chunk_id
.is_set
);
2610 stream
->current_chunk_id
.value
= stream_info
.new_chunk_id
;
2612 if (stream
->is_metadata
) {
2614 * Metadata streams have no index; consider its rotation
2617 stream
->index_rotated
= true;
2619 * The metadata stream is sent only over the control connection
2620 * so we know we have all the data to perform the stream
2623 ret
= do_rotate_stream_data(stream
);
2625 stream
->rotate_at_seq_num
= stream_info
.rotate_at_seq_num
;
2626 ret
= try_rotate_stream_data(stream
);
2628 goto end_stream_unlock
;
2631 ret
= try_rotate_stream_index(stream
);
2633 goto end_stream_unlock
;
2638 pthread_mutex_unlock(&stream
->lock
);
2641 memset(&reply
, 0, sizeof(reply
));
2643 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2645 reply
.ret_code
= htobe32(LTTNG_OK
);
2647 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2648 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2649 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2650 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2660 * relay_mkdir: Create a folder on the disk.
2662 static int relay_mkdir(const struct lttcomm_relayd_hdr
*recv_hdr
,
2663 struct relay_connection
*conn
,
2664 const struct lttng_buffer_view
*payload
)
2667 struct relay_session
*session
= conn
->session
;
2668 struct lttcomm_relayd_mkdir path_info_header
;
2669 struct lttcomm_relayd_generic_reply reply
;
2673 struct lttng_buffer_view path_view
;
2675 if (!session
|| !conn
->version_check_done
) {
2676 ERR("Trying to create a directory before version check");
2678 goto end_no_session
;
2681 if (session
->major
== 2 && session
->minor
< 11) {
2683 * This client is not supposed to use this command since
2684 * it predates its introduction.
2686 ERR("relay_mkdir command is unsupported before LTTng 2.11");
2688 goto end_no_session
;
2691 header_len
= sizeof(path_info_header
);
2692 if (payload
->size
< header_len
) {
2693 ERR("Unexpected payload size in \"relay_mkdir\": expected >= %zu bytes, got %zu bytes",
2694 header_len
, payload
->size
);
2696 goto end_no_session
;
2699 memcpy(&path_info_header
, payload
->data
, header_len
);
2701 path_info_header
.length
= be32toh(path_info_header
.length
);
2703 if (payload
->size
< header_len
+ path_info_header
.length
) {
2704 ERR("Unexpected payload size in \"relay_mkdir\" including path: expected >= %zu bytes, got %zu bytes",
2705 header_len
+ path_info_header
.length
, payload
->size
);
2707 goto end_no_session
;
2710 /* Ensure that it fits in local path length. */
2711 if (path_info_header
.length
>= LTTNG_PATH_MAX
) {
2712 ret
= -ENAMETOOLONG
;
2713 ERR("Path name argument of mkdir command (%" PRIu32
" bytes) exceeds the maximal length allowed (%d bytes)",
2714 path_info_header
.length
, LTTNG_PATH_MAX
);
2718 path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2719 path_info_header
.length
);
2721 path
= create_output_path(path_view
.data
);
2723 ERR("Failed to create output path");
2728 DBG("MKDIR command has path \"%s\", changed to \"%s\"", path_view
.data
, path
);
2729 ret
= utils_mkdir_recursive(path
, S_IRWXU
| S_IRWXG
, -1, -1);
2731 ERR("relay creating output directory");
2738 memset(&reply
, 0, sizeof(reply
));
2740 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2742 reply
.ret_code
= htobe32(LTTNG_OK
);
2744 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2745 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2746 ERR("Failed to send \"mkdir\" command reply (ret = %zd)", send_ret
);
2755 static int validate_rotate_rename_path_length(const char *path_type
,
2756 uint32_t path_length
)
2760 if (path_length
> LTTNG_PATH_MAX
) {
2761 ret
= -ENAMETOOLONG
;
2762 ERR("rotate rename \"%s\" path name length (%" PRIu32
" bytes) exceeds the allowed size of %i bytes",
2763 path_type
, path_length
, LTTNG_PATH_MAX
);
2764 } else if (path_length
== 0) {
2766 ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type
);
2772 * relay_rotate_rename: rename the trace folder after a rotation is
2773 * completed. We are not closing any fd here, just moving the folder, so it
2774 * works even if data is still in-flight.
2776 static int relay_rotate_rename(const struct lttcomm_relayd_hdr
*recv_hdr
,
2777 struct relay_connection
*conn
,
2778 const struct lttng_buffer_view
*payload
)
2782 struct relay_session
*session
= conn
->session
;
2783 struct lttcomm_relayd_generic_reply reply
;
2784 struct lttcomm_relayd_rotate_rename header
;
2786 size_t received_paths_size
;
2787 char *complete_old_path
= NULL
, *complete_new_path
= NULL
;
2788 struct lttng_buffer_view old_path_view
;
2789 struct lttng_buffer_view new_path_view
;
2791 if (!session
|| !conn
->version_check_done
) {
2792 ERR("Trying to rename a trace folder before version check");
2797 if (session
->major
== 2 && session
->minor
< 11) {
2798 ERR("relay_rotate_rename command is unsupported before LTTng 2.11");
2803 header_len
= sizeof(header
);
2804 if (payload
->size
< header_len
) {
2805 ERR("Unexpected payload size in \"relay_rotate_rename\": expected >= %zu bytes, got %zu bytes",
2806 header_len
, payload
->size
);
2811 memcpy(&header
, payload
->data
, header_len
);
2813 header
.old_path_length
= be32toh(header
.old_path_length
);
2814 header
.new_path_length
= be32toh(header
.new_path_length
);
2815 received_paths_size
= header
.old_path_length
+ header
.new_path_length
;
2817 if (payload
->size
< header_len
+ received_paths_size
) {
2818 ERR("Unexpected payload size in \"relay_rotate_rename\" including paths: expected >= %zu bytes, got %zu bytes",
2819 header_len
, payload
->size
);
2824 /* Ensure the paths don't exceed their allowed size. */
2825 ret
= validate_rotate_rename_path_length("old", header
.old_path_length
);
2829 ret
= validate_rotate_rename_path_length("new", header
.new_path_length
);
2834 old_path_view
= lttng_buffer_view_from_view(payload
, header_len
,
2835 header
.old_path_length
);
2836 new_path_view
= lttng_buffer_view_from_view(payload
,
2837 header_len
+ header
.old_path_length
,
2838 header
.new_path_length
);
2840 /* Validate that both paths received are NULL terminated. */
2841 if (old_path_view
.data
[old_path_view
.size
- 1] != '\0') {
2842 ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)");
2846 if (new_path_view
.data
[new_path_view
.size
- 1] != '\0') {
2847 ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)");
2852 DBG("ROTATE_RENAME command has argument old path = \"%s\", new_path = \"%s\"",
2853 old_path_view
.data
, new_path_view
.data
);
2854 complete_old_path
= create_output_path(old_path_view
.data
);
2855 if (!complete_old_path
) {
2856 ERR("Failed to build old output path in rotate_rename command");
2861 complete_new_path
= create_output_path(new_path_view
.data
);
2862 if (!complete_new_path
) {
2863 ERR("Failed to build new output path in rotate_rename command");
2867 DBG("Expanded ROTATE_RENAME arguments to old path = \"%s\", new_path = \"%s\"",
2868 complete_old_path
, complete_new_path
);
2870 ret
= utils_mkdir_recursive(complete_new_path
, S_IRWXU
| S_IRWXG
,
2873 ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"",
2879 * If a domain has not yet created its channel, the domain-specific
2880 * folder might not exist, but this is not an error.
2882 ret
= rename(complete_old_path
, complete_new_path
);
2883 if (ret
< 0 && errno
!= ENOENT
) {
2884 PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"",
2885 complete_old_path
, complete_new_path
);
2891 memset(&reply
, 0, sizeof(reply
));
2893 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2895 reply
.ret_code
= htobe32(LTTNG_OK
);
2897 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2899 if (send_ret
< sizeof(reply
)) {
2900 ERR("Failed to send \"rotate rename\" command reply (ret = %zd)",
2906 free(complete_old_path
);
2907 free(complete_new_path
);
2912 * Check if all the streams in the session have completed the last rotation.
2913 * The chunk_id value is used to distinguish the cases where a stream was
2914 * closed on the consumerd before the rotation started but it still active on
2915 * the relayd, and the case where a stream appeared on the consumerd/relayd
2916 * after the last rotation started (in that case, it is already writing in the
2917 * new chunk folder).
2920 int relay_rotate_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2921 struct relay_connection
*conn
,
2922 const struct lttng_buffer_view
*payload
)
2924 struct relay_session
*session
= conn
->session
;
2925 struct lttcomm_relayd_rotate_pending msg
;
2926 struct lttcomm_relayd_rotate_pending_reply reply
;
2927 struct lttng_ht_iter iter
;
2928 struct relay_stream
*stream
;
2932 bool rotate_pending
= false;
2934 DBG("Rotate pending command received");
2936 if (!session
|| !conn
->version_check_done
) {
2937 ERR("Trying to check for data before version check");
2942 if (session
->major
== 2 && session
->minor
< 11) {
2943 ERR("Unsupported feature before 2.11");
2948 if (payload
->size
< sizeof(msg
)) {
2949 ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes",
2950 sizeof(msg
), payload
->size
);
2955 memcpy(&msg
, payload
->data
, sizeof(msg
));
2957 chunk_id
= be64toh(msg
.chunk_id
);
2959 DBG("Evaluating rotate pending for session \"%s\" and chunk id %" PRIu64
,
2960 session
->session_name
, chunk_id
);
2963 * Iterate over all the streams in the session and check if they are
2964 * still waiting for data to perform their rotation.
2967 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2969 if (!stream_get(stream
)) {
2972 if (stream
->trace
->session
!= session
) {
2976 pthread_mutex_lock(&stream
->lock
);
2977 if (stream
->rotate_at_seq_num
!= -1ULL) {
2978 /* We have not yet performed the rotation. */
2979 rotate_pending
= true;
2980 DBG("Stream %" PRIu64
" is still rotating",
2981 stream
->stream_handle
);
2982 } else if (stream
->current_chunk_id
.value
<= chunk_id
) {
2984 * Stream closed on the consumer but still active on the
2987 rotate_pending
= true;
2988 DBG("Stream %" PRIu64
" did not exist on the consumer "
2989 "when the last rotation started, but is"
2990 "still waiting for data before getting"
2992 stream
->stream_handle
);
2994 pthread_mutex_unlock(&stream
->lock
);
2996 if (rotate_pending
) {
3003 memset(&reply
, 0, sizeof(reply
));
3004 reply
.generic
.ret_code
= htobe32((uint32_t) LTTNG_OK
);
3005 reply
.is_pending
= (uint8_t) !!rotate_pending
;
3006 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
3008 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3009 ERR("Failed to send \"rotate pending\" command reply (ret = %zd)",
3018 #define DBG_CMD(cmd_name, conn) \
3019 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3021 static int relay_process_control_command(struct relay_connection
*conn
,
3022 const struct lttcomm_relayd_hdr
*header
,
3023 const struct lttng_buffer_view
*payload
)
3027 switch (header
->cmd
) {
3028 case RELAYD_CREATE_SESSION
:
3029 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
3030 ret
= relay_create_session(header
, conn
, payload
);
3032 case RELAYD_ADD_STREAM
:
3033 DBG_CMD("RELAYD_ADD_STREAM", conn
);
3034 ret
= relay_add_stream(header
, conn
, payload
);
3036 case RELAYD_START_DATA
:
3037 DBG_CMD("RELAYD_START_DATA", conn
);
3038 ret
= relay_start(header
, conn
, payload
);
3040 case RELAYD_SEND_METADATA
:
3041 DBG_CMD("RELAYD_SEND_METADATA", conn
);
3042 ret
= relay_recv_metadata(header
, conn
, payload
);
3044 case RELAYD_VERSION
:
3045 DBG_CMD("RELAYD_VERSION", conn
);
3046 ret
= relay_send_version(header
, conn
, payload
);
3048 case RELAYD_CLOSE_STREAM
:
3049 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
3050 ret
= relay_close_stream(header
, conn
, payload
);
3052 case RELAYD_DATA_PENDING
:
3053 DBG_CMD("RELAYD_DATA_PENDING", conn
);
3054 ret
= relay_data_pending(header
, conn
, payload
);
3056 case RELAYD_QUIESCENT_CONTROL
:
3057 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
3058 ret
= relay_quiescent_control(header
, conn
, payload
);
3060 case RELAYD_BEGIN_DATA_PENDING
:
3061 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
3062 ret
= relay_begin_data_pending(header
, conn
, payload
);
3064 case RELAYD_END_DATA_PENDING
:
3065 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
3066 ret
= relay_end_data_pending(header
, conn
, payload
);
3068 case RELAYD_SEND_INDEX
:
3069 DBG_CMD("RELAYD_SEND_INDEX", conn
);
3070 ret
= relay_recv_index(header
, conn
, payload
);
3072 case RELAYD_STREAMS_SENT
:
3073 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
3074 ret
= relay_streams_sent(header
, conn
, payload
);
3076 case RELAYD_RESET_METADATA
:
3077 DBG_CMD("RELAYD_RESET_METADATA", conn
);
3078 ret
= relay_reset_metadata(header
, conn
, payload
);
3080 case RELAYD_ROTATE_STREAM
:
3081 DBG_CMD("RELAYD_ROTATE_STREAM", conn
);
3082 ret
= relay_rotate_session_stream(header
, conn
, payload
);
3084 case RELAYD_ROTATE_RENAME
:
3085 DBG_CMD("RELAYD_ROTATE_RENAME", conn
);
3086 ret
= relay_rotate_rename(header
, conn
, payload
);
3088 case RELAYD_ROTATE_PENDING
:
3089 DBG_CMD("RELAYD_ROTATE_PENDING", conn
);
3090 ret
= relay_rotate_pending(header
, conn
, payload
);
3093 DBG_CMD("RELAYD_MKDIR", conn
);
3094 ret
= relay_mkdir(header
, conn
, payload
);
3096 case RELAYD_UPDATE_SYNC_INFO
:
3098 ERR("Received unknown command (%u)", header
->cmd
);
3099 relay_unknown_command(conn
);
3108 static enum relay_connection_status
relay_process_control_receive_payload(
3109 struct relay_connection
*conn
)
3112 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3113 struct lttng_dynamic_buffer
*reception_buffer
=
3114 &conn
->protocol
.ctrl
.reception_buffer
;
3115 struct ctrl_connection_state_receive_payload
*state
=
3116 &conn
->protocol
.ctrl
.state
.receive_payload
;
3117 struct lttng_buffer_view payload_view
;
3119 if (state
->left_to_receive
== 0) {
3120 /* Short-circuit for payload-less commands. */
3121 goto reception_complete
;
3124 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3125 reception_buffer
->data
+ state
->received
,
3126 state
->left_to_receive
, MSG_DONTWAIT
);
3128 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3129 PERROR("Unable to receive command payload on sock %d",
3131 status
= RELAY_CONNECTION_STATUS_ERROR
;
3134 } else if (ret
== 0) {
3135 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3136 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3141 assert(ret
<= state
->left_to_receive
);
3143 state
->left_to_receive
-= ret
;
3144 state
->received
+= ret
;
3146 if (state
->left_to_receive
> 0) {
3148 * Can't transition to the protocol's next state, wait to
3149 * receive the rest of the header.
3151 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3152 state
->received
, state
->left_to_receive
,
3158 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3159 conn
->sock
->fd
, state
->received
);
3161 * The payload required to process the command has been received.
3162 * A view to the reception buffer is forwarded to the various
3163 * commands and the state of the control is reset on success.
3165 * Commands are responsible for sending their reply to the peer.
3167 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3169 ret
= relay_process_control_command(conn
,
3170 &state
->header
, &payload_view
);
3172 status
= RELAY_CONNECTION_STATUS_ERROR
;
3176 ret
= connection_reset_protocol_state(conn
);
3178 status
= RELAY_CONNECTION_STATUS_ERROR
;
3184 static enum relay_connection_status
relay_process_control_receive_header(
3185 struct relay_connection
*conn
)
3188 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3189 struct lttcomm_relayd_hdr header
;
3190 struct lttng_dynamic_buffer
*reception_buffer
=
3191 &conn
->protocol
.ctrl
.reception_buffer
;
3192 struct ctrl_connection_state_receive_header
*state
=
3193 &conn
->protocol
.ctrl
.state
.receive_header
;
3195 assert(state
->left_to_receive
!= 0);
3197 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3198 reception_buffer
->data
+ state
->received
,
3199 state
->left_to_receive
, MSG_DONTWAIT
);
3201 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3202 PERROR("Unable to receive control command header on sock %d",
3204 status
= RELAY_CONNECTION_STATUS_ERROR
;
3207 } else if (ret
== 0) {
3208 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3209 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3214 assert(ret
<= state
->left_to_receive
);
3216 state
->left_to_receive
-= ret
;
3217 state
->received
+= ret
;
3219 if (state
->left_to_receive
> 0) {
3221 * Can't transition to the protocol's next state, wait to
3222 * receive the rest of the header.
3224 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3225 state
->received
, state
->left_to_receive
,
3230 /* Transition to next state: receiving the command's payload. */
3231 conn
->protocol
.ctrl
.state_id
=
3232 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3233 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3234 header
.circuit_id
= be64toh(header
.circuit_id
);
3235 header
.data_size
= be64toh(header
.data_size
);
3236 header
.cmd
= be32toh(header
.cmd
);
3237 header
.cmd_version
= be32toh(header
.cmd_version
);
3238 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3239 &header
, sizeof(header
));
3241 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3242 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
3245 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3246 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3248 status
= RELAY_CONNECTION_STATUS_ERROR
;
3252 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3254 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3255 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3258 status
= RELAY_CONNECTION_STATUS_ERROR
;
3262 if (header
.data_size
== 0) {
3264 * Manually invoke the next state as the poll loop
3265 * will not wake-up to allow us to proceed further.
3267 status
= relay_process_control_receive_payload(conn
);
3274 * Process the commands received on the control socket
3276 static enum relay_connection_status
relay_process_control(
3277 struct relay_connection
*conn
)
3279 enum relay_connection_status status
;
3281 switch (conn
->protocol
.ctrl
.state_id
) {
3282 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3283 status
= relay_process_control_receive_header(conn
);
3285 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3286 status
= relay_process_control_receive_payload(conn
);
3289 ERR("Unknown control connection protocol state encountered.");
3297 * Handle index for a data stream.
3299 * Called with the stream lock held.
3301 * Return 0 on success else a negative value.
3303 static int handle_index_data(struct relay_stream
*stream
, uint64_t net_seq_num
,
3304 bool rotate_index
, bool *flushed
, uint64_t total_size
)
3307 uint64_t data_offset
;
3308 struct relay_index
*index
;
3310 /* Get data offset because we are about to update the index. */
3311 data_offset
= htobe64(stream
->tracefile_size_current
);
3313 DBG("handle_index_data: stream %" PRIu64
" net_seq_num %" PRIu64
" data offset %" PRIu64
,
3314 stream
->stream_handle
, net_seq_num
, stream
->tracefile_size_current
);
3317 * Lookup for an existing index for that stream id/sequence
3318 * number. If it exists, the control thread has already received the
3319 * data for it, thus we need to write it to disk.
3321 index
= relay_index_get_by_id_or_create(stream
, net_seq_num
);
3327 if (rotate_index
|| !stream
->index_file
) {
3328 const char *stream_path
;
3331 * The data connection creates the stream's first index file.
3333 * This can happen _after_ a ROTATE_STREAM command. In
3334 * other words, the data of the first packet of this stream
3335 * can be received after a ROTATE_STREAM command.
3337 * The ROTATE_STREAM command changes the stream's path_name
3338 * to point to the "next" chunk. If a rotation is pending for
3339 * this stream, as indicated by "rotate_at_seq_num != -1ULL",
3340 * it means that we are still receiving data that belongs in the
3341 * stream's former path.
3343 * In this very specific case, we must ensure that the index
3344 * file is created in the streams's former path,
3347 * All other rotations beyond the first one are not affected
3348 * by this problem since the actual rotation operation creates
3349 * the new chunk's index file.
3351 stream_path
= stream
->rotate_at_seq_num
== -1ULL ?
3353 stream
->prev_path_name
;
3355 ret
= create_rotate_index_file(stream
, stream_path
);
3357 ERR("Failed to rotate index");
3358 /* Put self-ref for this index due to error. */
3359 relay_index_put(index
);
3365 if (relay_index_set_file(index
, stream
->index_file
, data_offset
)) {
3367 /* Put self-ref for this index due to error. */
3368 relay_index_put(index
);
3373 ret
= relay_index_try_flush(index
);
3375 tracefile_array_commit_seq(stream
->tfa
);
3376 stream
->index_received_seqcount
++;
3378 } else if (ret
> 0) {
3379 index
->total_size
= total_size
;
3386 * relay_index_try_flush is responsible for the self-reference
3387 * put of the index object on error.
3389 ERR("relay_index_try_flush error %d", ret
);
3396 static enum relay_connection_status
relay_process_data_receive_header(
3397 struct relay_connection
*conn
)
3400 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3401 struct data_connection_state_receive_header
*state
=
3402 &conn
->protocol
.data
.state
.receive_header
;
3403 struct lttcomm_relayd_data_hdr header
;
3404 struct relay_stream
*stream
;
3406 assert(state
->left_to_receive
!= 0);
3408 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3409 state
->header_reception_buffer
+ state
->received
,
3410 state
->left_to_receive
, MSG_DONTWAIT
);
3412 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3413 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3414 status
= RELAY_CONNECTION_STATUS_ERROR
;
3417 } else if (ret
== 0) {
3418 /* Orderly shutdown. Not necessary to print an error. */
3419 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3420 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3425 assert(ret
<= state
->left_to_receive
);
3427 state
->left_to_receive
-= ret
;
3428 state
->received
+= ret
;
3430 if (state
->left_to_receive
> 0) {
3432 * Can't transition to the protocol's next state, wait to
3433 * receive the rest of the header.
3435 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3436 state
->received
, state
->left_to_receive
,
3441 /* Transition to next state: receiving the payload. */
3442 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3444 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3445 header
.circuit_id
= be64toh(header
.circuit_id
);
3446 header
.stream_id
= be64toh(header
.stream_id
);
3447 header
.data_size
= be32toh(header
.data_size
);
3448 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3449 header
.padding_size
= be32toh(header
.padding_size
);
3450 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3452 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3454 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3455 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3457 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3458 conn
->sock
->fd
, header
.circuit_id
,
3459 header
.stream_id
, header
.data_size
,
3460 header
.net_seq_num
, header
.padding_size
);
3462 stream
= stream_get_by_id(header
.stream_id
);
3464 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3466 /* Protocol error. */
3467 status
= RELAY_CONNECTION_STATUS_ERROR
;
3471 pthread_mutex_lock(&stream
->lock
);
3473 /* Check if a rotation is needed. */
3474 if (stream
->tracefile_size
> 0 &&
3475 (stream
->tracefile_size_current
+ header
.data_size
) >
3476 stream
->tracefile_size
) {
3477 uint64_t old_id
, new_id
;
3479 old_id
= tracefile_array_get_file_index_head(stream
->tfa
);
3480 tracefile_array_file_rotate(stream
->tfa
);
3482 /* new_id is updated by utils_rotate_stream_file. */
3485 ret
= utils_rotate_stream_file(stream
->path_name
,
3486 stream
->channel_name
, stream
->tracefile_size
,
3487 stream
->tracefile_count
, -1,
3488 -1, stream
->stream_fd
->fd
,
3489 &new_id
, &stream
->stream_fd
->fd
);
3491 ERR("Failed to rotate stream output file");
3492 status
= RELAY_CONNECTION_STATUS_ERROR
;
3493 goto end_stream_unlock
;
3497 * Reset current size because we just performed a stream
3500 stream
->tracefile_size_current
= 0;
3501 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= true;
3505 pthread_mutex_unlock(&stream
->lock
);
3511 static enum relay_connection_status
relay_process_data_receive_payload(
3512 struct relay_connection
*conn
)
3515 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3516 struct relay_stream
*stream
;
3517 struct data_connection_state_receive_payload
*state
=
3518 &conn
->protocol
.data
.state
.receive_payload
;
3519 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3520 char data_buffer
[chunk_size
];
3521 bool partial_recv
= false;
3522 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3523 uint64_t left_to_receive
= state
->left_to_receive
;
3524 struct relay_session
*session
;
3526 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3527 state
->header
.stream_id
, state
->header
.net_seq_num
,
3528 state
->received
, left_to_receive
);
3530 stream
= stream_get_by_id(state
->header
.stream_id
);
3532 /* Protocol error. */
3533 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3534 state
->header
.stream_id
);
3535 status
= RELAY_CONNECTION_STATUS_ERROR
;
3539 pthread_mutex_lock(&stream
->lock
);
3540 session
= stream
->trace
->session
;
3541 if (!conn
->session
) {
3542 ret
= connection_set_session(conn
, session
);
3544 status
= RELAY_CONNECTION_STATUS_ERROR
;
3545 goto end_stream_unlock
;
3550 * The size of the "chunk" received on any iteration is bounded by:
3551 * - the data left to receive,
3552 * - the data immediately available on the socket,
3553 * - the on-stack data buffer
3555 while (left_to_receive
> 0 && !partial_recv
) {
3557 size_t recv_size
= min(left_to_receive
, chunk_size
);
3559 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3560 recv_size
, MSG_DONTWAIT
);
3562 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3563 PERROR("Socket %d error", conn
->sock
->fd
);
3564 status
= RELAY_CONNECTION_STATUS_ERROR
;
3566 goto end_stream_unlock
;
3567 } else if (ret
== 0) {
3568 /* No more data ready to be consumed on socket. */
3569 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3570 state
->header
.stream_id
);
3571 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3573 } else if (ret
< (int) recv_size
) {
3575 * All the data available on the socket has been
3578 partial_recv
= true;
3583 /* Write data to stream output fd. */
3584 write_ret
= lttng_write(stream
->stream_fd
->fd
, data_buffer
,
3586 if (write_ret
< (ssize_t
) recv_size
) {
3587 ERR("Relay error writing data to file");
3588 status
= RELAY_CONNECTION_STATUS_ERROR
;
3589 goto end_stream_unlock
;
3592 left_to_receive
-= recv_size
;
3593 state
->received
+= recv_size
;
3594 state
->left_to_receive
= left_to_receive
;
3596 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64
,
3597 write_ret
, stream
->stream_handle
);
3600 if (state
->left_to_receive
> 0) {
3602 * Did not receive all the data expected, wait for more data to
3603 * become available on the socket.
3605 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3606 state
->header
.stream_id
, state
->received
,
3607 state
->left_to_receive
);
3608 goto end_stream_unlock
;
3611 ret
= write_padding_to_file(stream
->stream_fd
->fd
,
3612 state
->header
.padding_size
);
3613 if ((int64_t) ret
< (int64_t) state
->header
.padding_size
) {
3614 ERR("write_padding_to_file: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3615 stream
->stream_handle
,
3616 state
->header
.net_seq_num
, ret
);
3617 status
= RELAY_CONNECTION_STATUS_ERROR
;
3618 goto end_stream_unlock
;
3622 if (session_streams_have_index(session
)) {
3623 ret
= handle_index_data(stream
, state
->header
.net_seq_num
,
3624 state
->rotate_index
, &index_flushed
, state
->header
.data_size
+ state
->header
.padding_size
);
3626 ERR("handle_index_data: fail stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3627 stream
->stream_handle
,
3628 state
->header
.net_seq_num
, ret
);
3629 status
= RELAY_CONNECTION_STATUS_ERROR
;
3630 goto end_stream_unlock
;
3634 stream
->tracefile_size_current
+= state
->header
.data_size
+
3635 state
->header
.padding_size
;
3637 if (stream
->prev_data_seq
== -1ULL) {
3640 if (index_flushed
) {
3641 stream
->pos_after_last_complete_data_index
=
3642 stream
->tracefile_size_current
;
3643 stream
->prev_index_seq
= state
->header
.net_seq_num
;
3644 ret
= try_rotate_stream_index(stream
);
3646 goto end_stream_unlock
;
3650 stream
->prev_data_seq
= state
->header
.net_seq_num
;
3653 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3654 * contents of *state which are aliased (union) to the same location as
3655 * the new state. Don't use it beyond this point.
3657 connection_reset_protocol_state(conn
);
3660 ret
= try_rotate_stream_data(stream
);
3662 status
= RELAY_CONNECTION_STATUS_ERROR
;
3663 goto end_stream_unlock
;
3667 close_requested
= stream
->close_requested
;
3668 pthread_mutex_unlock(&stream
->lock
);
3669 if (close_requested
&& left_to_receive
== 0) {
3670 try_stream_close(stream
);
3674 pthread_mutex_lock(&session
->lock
);
3675 uatomic_set(&session
->new_streams
, 1);
3676 pthread_mutex_unlock(&session
->lock
);
3685 * relay_process_data: Process the data received on the data socket
3687 static enum relay_connection_status
relay_process_data(
3688 struct relay_connection
*conn
)
3690 enum relay_connection_status status
;
3692 switch (conn
->protocol
.data
.state_id
) {
3693 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3694 status
= relay_process_data_receive_header(conn
);
3696 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3697 status
= relay_process_data_receive_payload(conn
);
3700 ERR("Unexpected data connection communication state.");
3707 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3711 (void) lttng_poll_del(events
, pollfd
);
3713 ret
= close(pollfd
);
3715 ERR("Closing pollfd %d", pollfd
);
3719 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3720 int pollfd
, struct relay_connection
*conn
)
3722 const char *type_str
;
3724 switch (conn
->type
) {
3729 type_str
= "Control";
3731 case RELAY_VIEWER_COMMAND
:
3732 type_str
= "Viewer Command";
3734 case RELAY_VIEWER_NOTIFICATION
:
3735 type_str
= "Viewer Notification";
3738 type_str
= "Unknown";
3740 cleanup_connection_pollfd(events
, pollfd
);
3741 connection_put(conn
);
3742 DBG("%s connection closed with %d", type_str
, pollfd
);
3746 * This thread does the actual work
3748 static void *relay_thread_worker(void *data
)
3750 int ret
, err
= -1, last_seen_data_fd
= -1;
3752 struct lttng_poll_event events
;
3753 struct lttng_ht
*relay_connections_ht
;
3754 struct lttng_ht_iter iter
;
3755 struct relay_connection
*destroy_conn
= NULL
;
3757 DBG("[thread] Relay worker started");
3759 rcu_register_thread();
3761 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3763 if (testpoint(relayd_thread_worker
)) {
3764 goto error_testpoint
;
3767 health_code_update();
3769 /* table of connections indexed on socket */
3770 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3771 if (!relay_connections_ht
) {
3772 goto relay_connections_ht_error
;
3775 ret
= create_thread_poll_set(&events
, 2);
3777 goto error_poll_create
;
3780 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3787 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3789 health_code_update();
3791 /* Infinite blocking call, waiting for transmission */
3792 DBG3("Relayd worker thread polling...");
3793 health_poll_entry();
3794 ret
= lttng_poll_wait(&events
, -1);
3798 * Restart interrupted system call.
3800 if (errno
== EINTR
) {
3809 * Process control. The control connection is
3810 * prioritized so we don't starve it with high
3811 * throughput tracing data on the data connection.
3813 for (i
= 0; i
< nb_fd
; i
++) {
3814 /* Fetch once the poll data */
3815 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3816 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3818 health_code_update();
3820 /* Thread quit pipe has been closed. Killing thread. */
3821 ret
= check_thread_quit_pipe(pollfd
, revents
);
3827 /* Inspect the relay conn pipe for new connection */
3828 if (pollfd
== relay_conn_pipe
[0]) {
3829 if (revents
& LPOLLIN
) {
3830 struct relay_connection
*conn
;
3832 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3836 lttng_poll_add(&events
, conn
->sock
->fd
,
3837 LPOLLIN
| LPOLLRDHUP
);
3838 connection_ht_add(relay_connections_ht
, conn
);
3839 DBG("Connection socket %d added", conn
->sock
->fd
);
3840 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3841 ERR("Relay connection pipe error");
3844 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
3848 struct relay_connection
*ctrl_conn
;
3850 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3851 /* If not found, there is a synchronization issue. */
3854 if (ctrl_conn
->type
== RELAY_DATA
) {
3855 if (revents
& LPOLLIN
) {
3857 * Flag the last seen data fd not deleted. It will be
3858 * used as the last seen fd if any fd gets deleted in
3861 last_notdel_data_fd
= pollfd
;
3863 goto put_ctrl_connection
;
3865 assert(ctrl_conn
->type
== RELAY_CONTROL
);
3867 if (revents
& LPOLLIN
) {
3868 enum relay_connection_status status
;
3870 status
= relay_process_control(ctrl_conn
);
3871 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3873 * On socket error flag the session as aborted to force
3874 * the cleanup of its stream otherwise it can leak
3875 * during the lifetime of the relayd.
3877 * This prevents situations in which streams can be
3878 * left opened because an index was received, the
3879 * control connection is closed, and the data
3880 * connection is closed (uncleanly) before the packet's
3883 * Since the control connection encountered an error,
3884 * it is okay to be conservative and close the
3885 * session right now as we can't rely on the protocol
3886 * being respected anymore.
3888 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3889 session_abort(ctrl_conn
->session
);
3892 /* Clear the connection on error or close. */
3893 relay_thread_close_connection(&events
,
3898 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3899 relay_thread_close_connection(&events
,
3901 if (last_seen_data_fd
== pollfd
) {
3902 last_seen_data_fd
= last_notdel_data_fd
;
3905 ERR("Unexpected poll events %u for control sock %d",
3907 connection_put(ctrl_conn
);
3910 put_ctrl_connection
:
3911 connection_put(ctrl_conn
);
3916 * The last loop handled a control request, go back to poll to make
3917 * sure we prioritise the control socket.
3923 if (last_seen_data_fd
>= 0) {
3924 for (i
= 0; i
< nb_fd
; i
++) {
3925 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3927 health_code_update();
3929 if (last_seen_data_fd
== pollfd
) {
3936 /* Process data connection. */
3937 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
3938 /* Fetch the poll data. */
3939 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3940 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3941 struct relay_connection
*data_conn
;
3943 health_code_update();
3946 /* No activity for this FD (poll implementation). */
3950 /* Skip the command pipe. It's handled in the first loop. */
3951 if (pollfd
== relay_conn_pipe
[0]) {
3955 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
3957 /* Skip it. Might be removed before. */
3960 if (data_conn
->type
== RELAY_CONTROL
) {
3961 goto put_data_connection
;
3963 assert(data_conn
->type
== RELAY_DATA
);
3965 if (revents
& LPOLLIN
) {
3966 enum relay_connection_status status
;
3968 status
= relay_process_data(data_conn
);
3969 /* Connection closed or error. */
3970 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
3972 * On socket error flag the session as aborted to force
3973 * the cleanup of its stream otherwise it can leak
3974 * during the lifetime of the relayd.
3976 * This prevents situations in which streams can be
3977 * left opened because an index was received, the
3978 * control connection is closed, and the data
3979 * connection is closed (uncleanly) before the packet's
3982 * Since the data connection encountered an error,
3983 * it is okay to be conservative and close the
3984 * session right now as we can't rely on the protocol
3985 * being respected anymore.
3987 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
3988 session_abort(data_conn
->session
);
3990 relay_thread_close_connection(&events
, pollfd
,
3993 * Every goto restart call sets the last seen fd where
3994 * here we don't really care since we gracefully
3995 * continue the loop after the connection is deleted.
3998 /* Keep last seen port. */
3999 last_seen_data_fd
= pollfd
;
4000 connection_put(data_conn
);
4003 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4004 relay_thread_close_connection(&events
, pollfd
,
4007 ERR("Unknown poll events %u for data sock %d",
4010 put_data_connection
:
4011 connection_put(data_conn
);
4013 last_seen_data_fd
= -1;
4016 /* Normal exit, no error */
4021 /* Cleanup remaining connection object. */
4023 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
4026 health_code_update();
4028 session_abort(destroy_conn
->session
);
4031 * No need to grab another ref, because we own
4034 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
4039 lttng_poll_clean(&events
);
4041 lttng_ht_destroy(relay_connections_ht
);
4042 relay_connections_ht_error
:
4043 /* Close relay conn pipes */
4044 utils_close_pipe(relay_conn_pipe
);
4046 DBG("Thread exited with error");
4048 DBG("Worker thread cleanup complete");
4052 ERR("Health error occurred in %s", __func__
);
4054 health_unregister(health_relayd
);
4055 rcu_unregister_thread();
4056 lttng_relay_stop_threads();
4061 * Create the relay command pipe to wake thread_manage_apps.
4062 * Closed in cleanup().
4064 static int create_relay_conn_pipe(void)
4068 ret
= utils_create_pipe_cloexec(relay_conn_pipe
);
4076 int main(int argc
, char **argv
)
4078 int ret
= 0, retval
= 0;
4081 /* Parse arguments */
4083 if (set_options(argc
, argv
)) {
4088 if (set_signal_handler()) {
4093 /* Try to create directory if -o, --output is specified. */
4094 if (opt_output_path
) {
4095 if (*opt_output_path
!= '/') {
4096 ERR("Please specify an absolute path for -o, --output PATH");
4101 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4104 ERR("Unable to create %s", opt_output_path
);
4111 if (opt_daemon
|| opt_background
) {
4114 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4122 * We are in the child. Make sure all other file
4123 * descriptors are closed, in case we are called with
4124 * more opened file descriptors than the standard ones.
4126 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
4131 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
4132 if (!sessiond_trace_chunk_registry
) {
4133 ERR("Failed to initialize session daemon trace chunk registry");
4135 goto exit_sessiond_trace_chunk_registry
;
4138 /* Initialize thread health monitoring */
4139 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4140 if (!health_relayd
) {
4141 PERROR("health_app_create error");
4143 goto exit_health_app_create
;
4146 /* Create thread quit pipe */
4147 if (init_thread_quit_pipe()) {
4149 goto exit_init_data
;
4152 /* Setup the thread apps communication pipe. */
4153 if (create_relay_conn_pipe()) {
4155 goto exit_init_data
;
4158 /* Init relay command queue. */
4159 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4161 /* Initialize communication library */
4163 lttcomm_inet_init();
4165 /* tables of sessions indexed by session ID */
4166 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4169 goto exit_init_data
;
4172 /* tables of streams indexed by stream ID */
4173 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4174 if (!relay_streams_ht
) {
4176 goto exit_init_data
;
4179 /* tables of streams indexed by stream ID */
4180 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4181 if (!viewer_streams_ht
) {
4183 goto exit_init_data
;
4186 ret
= utils_create_pipe(health_quit_pipe
);
4189 goto exit_health_quit_pipe
;
4192 /* Create thread to manage the client socket */
4193 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4194 thread_manage_health
, (void *) NULL
);
4197 PERROR("pthread_create health");
4199 goto exit_health_thread
;
4202 /* Setup the dispatcher thread */
4203 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4204 relay_thread_dispatcher
, (void *) NULL
);
4207 PERROR("pthread_create dispatcher");
4209 goto exit_dispatcher_thread
;
4212 /* Setup the worker thread */
4213 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4214 relay_thread_worker
, NULL
);
4217 PERROR("pthread_create worker");
4219 goto exit_worker_thread
;
4222 /* Setup the listener thread */
4223 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4224 relay_thread_listener
, (void *) NULL
);
4227 PERROR("pthread_create listener");
4229 goto exit_listener_thread
;
4232 ret
= relayd_live_create(live_uri
);
4234 ERR("Starting live viewer threads");
4240 * This is where we start awaiting program completion (e.g. through
4241 * signal that asks threads to teardown).
4244 ret
= relayd_live_join();
4250 ret
= pthread_join(listener_thread
, &status
);
4253 PERROR("pthread_join listener_thread");
4257 exit_listener_thread
:
4258 ret
= pthread_join(worker_thread
, &status
);
4261 PERROR("pthread_join worker_thread");
4266 ret
= pthread_join(dispatcher_thread
, &status
);
4269 PERROR("pthread_join dispatcher_thread");
4272 exit_dispatcher_thread
:
4274 ret
= pthread_join(health_thread
, &status
);
4277 PERROR("pthread_join health_thread");
4282 utils_close_pipe(health_quit_pipe
);
4283 exit_health_quit_pipe
:
4286 health_app_destroy(health_relayd
);
4287 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry
);
4288 exit_health_app_create
:
4289 exit_sessiond_trace_chunk_registry
:
4292 * Wait for all pending call_rcu work to complete before tearing
4293 * down data structures. call_rcu worker may be trying to
4294 * perform lookups in those structures.
4299 /* Ensure all prior call_rcu are done. */